In mid-June, the Norwegian government announced that primary school students would essentially stop using generative AI starting this fall. A few months earlier, in Syracuse, New York, the law firm Barclay Damon adopted an internal rule: For their first three years, junior lawyers must produce their work without AI, and only then check it against the tool.
A Nordic public school system and a 300-person American law firm aren’t pursuing the same goals. But they arrived at the same conclusion: Beginners must first learn without assistance in order to be, later on, well assisted.
Norway’s decision involves different stages
The word ban made the headlines, but Norway’s decision is more nuanced than that. From ages 6 to 13, students should, as a rule, not touch generative AI. From 14 to 16, they can use it cautiously, under a teacher’s supervision. From 17 to 19, they must learn to use it properly, to prepare for higher education and work.
“The most important thing in school is that our children learn to read, write, and count,” Prime Minister Jonas Gahr Støre summed up. The government also pledged to fund the return of paper books to classrooms—a reversal of 30 years of massive digital investment, in a country that adopted computers in schools as early as the 1990s and tablets by 2010.
Norway has run this experiment before. In 2024, it banned smartphones from schools in response to declining academic results. According to a study by the Norwegian Institute of Public Health, bullying decreased and grades went back up. Armed with that precedent, the government chooses to do it again with AI. Some foundational learning requires doing things yourself, slowly, with your own means.
The three-year rule
Barclay Damon transposes exactly the same logic into the workplace. At a New York State Bar Association panel, managing partner Connie Cahill laid out the firm’s policy: Every associate in their first three years must do their work without AI first. They turn in their draft, their outline, their memo—whatever the deliverable—and only then can they go verify and improve it with the tool. Doug Nash, who chairs the firm’s AI committee, explained why: Training the next generation of lawyers is already the hardest part of his job, and handing beginners a shortcut on day one deprives them of the path that builds professional judgment.
It’s worth measuring what this rule costs. A law firm bills by the hour. Every memo a junior spends three days drafting alone, when AI could produce an apparently good enough version in 10 minutes, is paid for in unoptimized hours. Barclay Damon chose to absorb that immediate efficiency loss because the firm needs, 10 years from now, senior partners capable of judging legal reasoning, including a machine’s.
Norway is making the same trade-off with public money: Buying books back and having children work unassisted means accepting a less “modern” school today in exchange for more solid adults tomorrow.
Learn first, delegate later
The order of operations matters. You learn your multiplication tables before you use a calculator. You log hours of supervised driving before taking the wheel alone. Civil aviation requires pilots to fly a mandatory number of manual hours, even when autopilot would do better, because the industry learned from several crashes what it costs to let skills atrophy.
For beginners, AI threatens the very foundations. A study by Microsoft Research and Carnegie Mellon showed that the more people trust AI, the less they exercise their critical thinking. Only those who are confident in their own skills dare question what the machine produces. In other words, to verify AI well, you need to have learned to work without it.
Those unassisted skills are also what you fall back on the day the tool goes away: a power outage, a discontinued service, a cyberattack on a data center, or a natural disaster. And climate change guarantees more of those.
Norwegian primary school students and first-year lawyers are in the same situation: They don’t yet have the necessary base. Giving them the tool too early removes the bottom rungs of the ladder and drains the pipeline of tomorrow’s experts.
Against resignation
The public debate about AI generally oscillates between two postures. On one side, calls for outright refusal, like this recent op-ed in French daily Le Monde urging a general boycott of generative AI. On the other, resignation: Since students will use it anyway, why bother resisting?
The examples of Norway and Barclay Damon point to a third way. Neither rejects the technology altogether. Norwegian high schoolers will learn to use it; fourth-year associates will use it daily. But both refuse resignation and decide to protect the learning phase, even if it means being less “efficient” in the short term.
This is an argument about robustness. As the French biologist Olivier Hamant argues, robust systems survive shocks precisely because they carry redundancies that look wasteful in normal times: two kidneys, backup metabolic pathways, spare capacity everywhere.
Nature is rarely optimized, and that is exactly what makes it resilient. People who know how to do the work without the machine are that redundancy, and an organization that optimizes them away becomes fragile by design.
Right now, organizations everywhere are rolling out training programs to get everyone up to speed on AI. It may be time to ask the opposite question: Where are the AI-free zones we need? The experts you’ll rely on in 10 years are the beginners you’re training, or not training, today.